Communication base station battery predictive maintenance method and system based on artificial intelligence

Abstract

The invention relates to the field of power management, in particular to a communication base station battery predictive maintenance method and system based on artificial intelligence, and the method comprises the steps: carrying out the data monitoring of an iron-lithium battery of a communication base station, so as to obtain original monitoring data; performing lithium iron battery health condition evaluation on the original monitoring data through a health supervision model of a pre-trained multi-model combination structure to obtain health evaluation information, drawing a lithium iron battery health change curve based on the health evaluation information at each moment, and performing doubtful point labeling and condition speculation on the health change curve to obtain a lithium iron battery health condition evaluation result; a plurality of speculative health curves are generated, the potential risk of the lithium iron battery is judged by combining the health change curve and each speculative health curve, and a corresponding maintenance scheme is allocated to the lithium iron battery according to a judgment result, so that predictive maintenance of the lithium iron battery of the communication base station can be realized, the service life of the battery is prolonged, and the risk of sudden failure of the battery is reduced. Therefore, the station power supply reliability is improved.

Description

Technical Field

[0001] This invention relates to the field of power management, specifically to a predictive maintenance method and system for communication base station batteries based on artificial intelligence. Background Technology

[0002] Modern society demands increasingly higher communication quality, such as clear voice calls and stable and fast data transmission. Poor performance of communication base station batteries can cause base stations to malfunction during power outages, affecting the quality of communication services and even causing communication interruptions. To meet users' growing demands for communication quality, it is essential to ensure that base station batteries are always in good working order. Traditional communication base station battery maintenance mainly relies on regular inspections and manual testing. This method not only consumes a lot of manpower, resources, and time, but also has limited testing frequency and accuracy, making it difficult to detect potential battery problems in a timely manner. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a predictive maintenance method and system for communication base station batteries based on artificial intelligence, which can solve the problems in the prior art.

[0004] This invention is achieved through the following technical solution: This invention provides a predictive maintenance method for communication base station batteries based on artificial intelligence, comprising: Data monitoring is performed on the lithium iron phosphate battery and communication module of the communication base station to obtain raw monitoring data; The health status of the lithium iron phosphate battery is assessed by a pre-trained health supervision model based on the raw monitoring data, and health assessment information is obtained. Based on the health assessment information at each time point, the health change curve of the lithium iron phosphate battery is plotted, and the suspicious points of the health change curve are marked and the condition is inferred to generate several inferred health curves. The potential risks of the lithium iron phosphate battery are assessed by combining the health change curves and the inferred health curves, and corresponding maintenance plans are assigned to the lithium iron phosphate battery based on the assessment results.

[0005] This invention provides an artificial intelligence-based predictive maintenance system for communication base station batteries, used to implement the artificial intelligence-based predictive maintenance method for communication base station batteries as described in any one of the first aspects, comprising: The data monitoring module is used to monitor the lithium iron phosphate battery and communication module of the communication base station to obtain raw monitoring data. The health assessment module is used to assess the health status of the lithium iron phosphate battery based on the raw monitoring data using a pre-trained health supervision model, and to obtain health assessment information. The suspicious point prediction module is used to draw the health change curve of the lithium iron phosphate battery based on the health assessment information at each time point, and to mark suspicious points and predict the condition of the health change curve to generate several predicted health curves. The risk maintenance module is used to assess the potential risks of the lithium iron phosphate battery by combining the health change curves and the inferred health curves, and to assign a corresponding maintenance plan to the lithium iron phosphate battery based on the assessment results.

[0006] In summary, the beneficial effects of this invention are: This invention provides comprehensive data monitoring, combining battery and communication module data to offer richer information for assessment. The health monitoring model provides accurate assessments, accurately judging battery health status based on raw data, plotting health change curves, marking suspicious points, and generating predictive curves. It can intuitively display changes in battery status, predict potential trends, and use comprehensive curves to identify potential risks and allocate maintenance plans. This allows for early detection of problems, precise maintenance, reduced operation and maintenance costs, and ensures stable operation of communication base station batteries. Attached Figure Description

[0007] For ease of explanation, the present invention will be described in detail below with reference to specific embodiments and accompanying drawings.

[0008] Figure 1 This is a schematic diagram illustrating the steps of a predictive maintenance method for communication base station batteries based on artificial intelligence, according to the present invention. Figure 2 This is a schematic diagram of the structure of a predictive maintenance system for communication base station batteries based on artificial intelligence, according to the present invention. Detailed Implementation

[0009] All features disclosed in this specification, or steps in all methods or processes disclosed herein, may be combined in any way, except for mutually exclusive features and / or steps.

[0010] The following is combined with Figure 1-2 The present invention will be described in detail below.

[0011] like Figure 1 As shown, this invention provides a predictive maintenance method for communication base station batteries based on artificial intelligence, comprising: S1: Perform data monitoring on the lithium iron phosphate battery and communication module of the communication base station to obtain raw monitoring data; S2: The health status of the lithium iron phosphate battery is assessed using a pre-trained health supervision model based on the original monitoring data to obtain health assessment information; S3: Based on the health assessment information at each time point, plot the health change curve of the lithium iron phosphate battery, mark the suspicious points and speculate on the condition of the health change curve, and generate several speculative health curves. S4: Combine the health change curves with the predicted health curves to determine the potential risks of the lithium iron phosphate battery, and allocate corresponding maintenance plans for the lithium iron phosphate battery based on the determination results.

[0012] In S1, various sensors are installed in the lithium iron phosphate battery, such as voltage sensors to measure the battery voltage, current sensors to monitor the current, and temperature sensors to obtain the battery's operating temperature. These sensors collect data in real time and transmit the data to a data acquisition device. The data acquisition device transmits the collected battery performance data to the monitoring system for storage and processing through specific communication interfaces (such as RS-485, CAN bus, etc.) and communication protocols. According to actual needs, an appropriate data recording frequency is set, such as recording data every few minutes or seconds, to ensure that changes in battery performance can be captured.

[0013] Battery parameters such as voltage, current, and temperature are key indicators reflecting their health status. For example, abnormal voltage indicates a short circuit or open circuit problem inside the battery; excessively high temperature will accelerate the aging process of the battery. By collecting these data, potential problems with the battery can be detected in time. By monitoring various performance data of the battery over a long period of time, the performance change trend of the battery can be analyzed, thereby predicting the remaining lifespan of the battery and providing a basis for battery replacement.

[0014] Network monitoring software or hardware devices can be used to monitor the communication quality of communication modules. For example, network analyzers can be used to measure signal strength, signal quality, and bit error rate; traffic monitoring tools can be used to statistically analyze the transmission rate and packet loss rate of communication data. Communication modules themselves usually record some log information during the communication process. By reading and analyzing these logs, communication quality-related data can be obtained. A real-time monitoring mechanism can be set up to promptly feed the collected communication quality data back to the monitoring system so as to detect communication anomalies in a timely manner.

[0015] Communication quality data such as signal strength and packet loss rate can intuitively reflect the working stability of the communication module. If the signal strength is too low or the packet loss rate is too high, it will lead to communication interruption or data transmission errors, affecting the normal operation of the communication base station. The performance of lithium iron phosphate batteries may affect the operation of the communication module, and vice versa. By collecting communication quality data, the correlation between battery performance and communication quality can be analyzed, providing a basis for comprehensively judging the operating status of the communication base station.

[0016] Develop or use specialized data integration software to integrate the collected battery performance data and communication quality data. The software will pair and combine battery data and communication data from the same point in time based on information such as the timestamp of the data, and store the combined data in a unified format, such as a database table or CSV file, for subsequent analysis and processing.

[0017] Individual battery performance data or communication quality data can only reflect one aspect of the communication base station's situation. However, by combining the two, a more comprehensive understanding of the overall operating status of the communication base station can be obtained. For example, when it is found that battery performance is deteriorating while communication quality is also problematic, it can be inferred that there may be some kind of correlation between the two, thus allowing for more in-depth analysis and investigation. Subsequent steps such as health status assessment and potential risk judgment need to comprehensively consider the condition of the battery and communication module. The completeness of the original monitoring data is crucial for improving the accuracy of assessment and judgment.

[0018] In S2, the raw monitoring data usually contains timestamps to identify the collection time of each data point. The timestamp information can be extracted from the data, and a suitable sorting algorithm, such as quicksort or mergesort, can be used to sort the raw monitoring data in ascending order according to the timestamps. Common programming languages ​​have built-in sorting functions that can be used directly. The sorted data can be used to form the raw monitoring sequence, which can be stored in data structures such as arrays and lists for easy subsequent processing.

[0019] The operating state of lithium iron phosphate batteries changes over time. By arranging the data in a time series, the temporal order of the data can be preserved, thereby better reflecting the evolution of battery performance over time. For example, by observing time series data, it can be found whether the battery performance has periodic changes or a gradual deterioration trend. Subsequent health monitoring models often need to consider the time series characteristics of the data. Sort the data by time so that the model can learn the time dependence of the data and improve the accuracy of the assessment.

[0020] Analyze the time characteristics of the current moment, such as weekdays, weekends, daytime, nighttime, and different seasons. These time characteristics may affect battery usage. For example, during the day, the traffic volume of communication base stations is higher, and the battery charging and discharging frequency will also be higher. Based on different time characteristics, pre-set corresponding information reference modes. For example, for the case of weekday daytime, the reference mode can be to take the current moment as the center and extract the data within one hour before and after; for the case of weekend night, the reference mode can extract the data within two hours before and after.

[0021] Following a pre-defined information reference pattern, several segments of specified data are extracted from the original monitoring sequence to obtain health reference information. These extracted sequences contain battery operating data over different time periods. Batteries exhibit different operating modes and performance characteristics at different times; for example, battery performance is affected during high-temperature seasons, and the number of charge-discharge cycles is higher during peak business periods. By determining the information reference pattern and extracting data based on time characteristics, data relevant to the current situation can be selected more accurately for evaluation, improving the reliability of the evaluation results. Extracting sequences of different specifications can provide health reference information for multiple time ranges, enabling the model to analyze battery health from multiple perspectives and avoiding the limitations of data from a single time period.

[0022] The pre-trained health supervision model is loaded into memory. This model can be trained based on machine learning algorithms (such as neural networks, decision trees, etc.). Before inputting health reference information into the model, the data needs to be preprocessed, such as normalized and standardized, to ensure that the data format and range meet the model's input requirements. The preprocessed health reference information is then input into the health supervision model, which calculates and analyzes it, outputting several health assessment features, such as battery health scores and failure probabilities.

[0023] The pre-trained health supervision model has learned the relationship between a large amount of battery data and the corresponding health status. It can extract valuable features from the input health reference information. Through the model's calculation and prediction, the health status of the battery can be assessed more objectively and accurately, avoiding the subjectivity and limitations of human judgment. The health assessment features can present the health status of the battery in a quantitative form, which is convenient for subsequent processing and analysis. For example, the health score can intuitively reflect the overall health level of the battery, making it easier for managers to quickly understand the status of the battery.

[0024] Calculate the correlation between different health assessment features, for example, using methods such as the Pearson correlation coefficient. Correlation analysis can reveal which features are strongly correlated and which are relatively independent. Based on the results of the correlation analysis and actual business needs, information verification rules are set. For example, if there is a strong positive correlation between two features, but in a certain assessment one feature shows good battery health while the other shows a serious fault, then it is necessary to further check the accuracy of the data or the reasonableness of the model's output. Appropriate information fusion methods should be selected, such as weighted average or evidence theory, to fuse the verified health assessment features. In the weighted average method, different weights can be assigned to each feature according to its importance, and then the values ​​of each feature are weighted and summed to obtain the final health assessment information.

[0025] Different health assessment features can reflect the health status of a battery from different perspectives, and there is a certain degree of error or uncertainty. Through interactive information verification, some erroneous or unreasonable feature values ​​can be identified and corrected. Information fusion can combine the advantages of each feature to improve the accuracy and reliability of the assessment results. Integrating various health assessment features into a unified health assessment information can provide a more comprehensive and concise basis for subsequent maintenance decisions. Managers can quickly judge the health status of the battery based on this comprehensive information and take corresponding maintenance measures.

[0026] In S3, a multi-dimensional information space is defined. The dimensions of this space can be set according to actual needs, such as time dimension, battery health dimension, different performance index dimensions, etc. Based on the dimensions of the information space and the characteristics of the health assessment information, mapping rules are formulated. For example, the time information in the health assessment information is mapped to the time dimension of the information space, and the battery health value is mapped to the health dimension. At each information level of the information space, the health assessment information is displayed according to the mapping rules. For example, at a specific time point and health level, the health assessment information at that moment is displayed in a visual way (such as color depth, icon size, etc.), thereby generating health feedback parameters for each information level.

[0027] The multidimensional nature of the information space can display the health status of the battery from multiple perspectives, helping maintenance personnel to have a more comprehensive understanding of the battery's performance in different aspects. For example, by combining the time dimension and different performance index dimensions, the changing trend of battery performance over time and the correlation between various indicators can be observed. The generated health feedback parameters provide basic data for the subsequent plotting of health change curves, enabling the curves to accurately reflect changes in the battery's health status.

[0028] The health feedback parameters of each information level are sorted in chronological order to ensure data continuity. Appropriate curve fitting methods, such as linear fitting, polynomial fitting, or spline fitting, are used to fit the sorted health feedback parameters. Through fitting, the hierarchical feedback curves of each information level can be obtained. The hierarchical feedback curves of each information level are integrated to form the health change curve of the lithium iron phosphate battery, which intuitively shows the change of battery health status over time.

[0029] Health change curves can clearly show the trend of battery health over a period of time, enabling maintenance personnel to quickly understand the battery's performance trajectory. For example, by observing the upward or downward trend of the curve, it can be determined whether the battery's health is improving or deteriorating. Hierarchical feedback curves at different information levels can be compared and analyzed, which helps to discover the performance differences and interrelationships of the battery in different aspects. For example, by comparing the hierarchical feedback curves of battery capacity and temperature, the impact of temperature on battery capacity can be analyzed.

[0030] The feedback curves at each level are analyzed to determine their trends, such as rising, falling, or stable. Simultaneously, the correlation between these trends is analyzed, such as whether some curves change synchronously or in opposite directions. Based on the trends of each part of the hierarchical feedback curves, statistical analysis methods (such as mean, standard deviation, and threshold judgment) are used to perform anomaly detection processing on the curves themselves, obtaining the first anomaly parameter. For example, if the curve value at a certain moment deviates from the mean by more than a certain standard deviation, then that point is considered potentially anomaly. Based on the correlation between the trends of each curve, the interaction between the feedback curves at each level is processed to obtain the second anomaly parameter. For example, if two previously related curves suddenly show divergent trends, then anomalies may exist. Combining the first and second anomaly parameters, suspicious points are identified in the feedback curves at each level, and these suspicious points are marked on the hierarchical feedback curves, such as using special symbols or colors.

[0031] By analyzing the curves side-by-side and detecting anomalies, abnormal points in the battery's health status can be identified. These anomalies indicate potential problems such as battery aging or hidden faults. Timely detection of these issues provides crucial clues for subsequent maintenance work. Marking suspicious points allows maintenance personnel to quickly pinpoint the problem, avoiding a comprehensive check of the entire curve and improving maintenance efficiency.

[0032] Using the feedback point of concern as a benchmark, historical and future information before and after the point of concern are selected on the health change curve. These are used as historical and developmental reference information for the feedback point of concern. Based on the historical reference information, statistical analysis methods or machine learning models are used to generate several speculative feedback parameters for the feedback point of concern. For example, the reasonable value range of the feedback point of concern can be predicted based on the changing patterns of historical data. The speculative feedback parameters can be used to replace the feedback point of concern. The future development of the speculative feedback parameters can be simulated based on the developmental reference information. Time series prediction models (such as ARIMA, LSTM, etc.) can be used to simulate the future development trend of battery health status, thereby obtaining the speculative health curves for each speculative feedback parameter.

[0033] Inferred health curves can help maintenance personnel predict the future health status of batteries and prepare for maintenance in advance. For example, if the inferred health curve shows that the battery health status will deteriorate rapidly, then battery replacement or repair can be arranged in a timely manner. Generating several inferred health curves can take into account different possibilities and provide multiple reference options for maintenance decisions. Maintenance personnel can formulate corresponding maintenance strategies based on the different inferred curves to reduce maintenance risks.

[0034] In S4, the numerical differences between the inferred health curve and the health change curve at the same time point are compared. For example, the difference between corresponding points of the two curves and the difference in slope can be calculated. The differences in the shape characteristics of the curves can also be analyzed, such as the differences in curvature, peak value, and valley value. The extracted curve difference features are combined into a vector according to certain rules. For example, the differences at different time points are arranged in sequence to form a vector. The differences in slope, curvature and other features of the curves are also added to the vector. The difference feature vectors of each inferred health curve and the health change curve are combined to form a difference feature matrix. Each row of the matrix represents a difference feature vector between the inferred health curve and the health change curve.

[0035] By extracting curve difference features and expressing them quantitatively, the differences between curves are transformed into specific values, facilitating subsequent calculations and analysis. The difference feature matrix can comprehensively reflect the differences between each inferred health curve and the health change curve. The difference feature matrix contains rich information that can be used to judge the potential risks of lithium iron phosphate batteries. For example, a large difference means that the battery has greater uncertainty and potential risks.

[0036] Establish a correlation model between the difference feature matrix and the operating status of each functional unit of the battery. This model can be based on machine learning algorithms, such as decision trees and neural networks, or it can be based on empirical rules. The difference feature matrix is ​​input into the correlation model, and the model infers based on the input features, outputting operational prediction information for each functional unit of the lithium iron phosphate battery. For example, it can predict whether the battery's charging module, discharging module, protection circuit, and other functional units are operating normally.

[0037] By analyzing the difference feature matrix, we can delve into the operating status of each functional unit of the battery from the overall curve differences. This helps to more accurately locate the parts of the battery that may have problems, providing a more specific direction for subsequent maintenance. The prediction of the operating status of each functional unit can detect potential fault hazards in advance and prevent the fault from developing further and causing the lithium iron phosphate battery to fail.

[0038] Collect other relevant information besides operational projections, such as historical battery fault records and environmental monitoring data. Integrate this multi-source information with the operational projections and use different verification methods to verify the operational projections. For example, consistency verification can be used to check whether information from different sources is consistent; logical verification can also be used to check whether the operational projections conform to the battery's working principles and logical rules. Based on the results of cross-validation, calculate the potential risk parameters for each functional unit of the lithium iron phosphate battery. For example, a risk score can be assigned to each functional unit as a potential risk parameter based on factors such as the degree of information consistency and the severity of abnormal situations.

[0039] Single operational predictions may contain errors or uncertainties. Cross-validation can integrate information from multiple sources, reduce errors, and improve the accuracy of potential risk assessment. Accurate potential risk parameters can help maintenance personnel develop maintenance plans more scientifically and allocate maintenance resources more rationally.

[0040] Retrieve testing records of lithium iron phosphate batteries from databases or other storage systems, including previous testing times and results. Based on factors such as battery type, usage, and potential risk parameters, formulate rules for scheduling the next round of testing. For example, for lithium iron phosphate batteries with higher potential risks, the testing cycle can be shortened; for lithium iron phosphate batteries in good operating condition, the testing cycle can be appropriately extended. Determine the next round of testing time for lithium iron phosphate batteries based on the testing records and the established rules.

[0041] By retrieving test records and scheduling the next round of testing in conjunction with the actual condition of the battery, over-testing or under-testing can be avoided. A reasonable testing plan can reduce testing costs while ensuring the safe operation of the battery. Regular testing can promptly identify new problems that arise during battery operation, ensuring that lithium iron phosphate batteries are always in good operating condition.

[0042] Establish an implementation value assessment model for software and hardware testing measures. This model can consider factors such as potential risk parameters, testing costs, and testing effects. For example, for functional units with high potential risks, testing measures with good testing effects but high costs have high implementation value. Input the potential risk parameters of each functional unit into the assessment model, calculate the implementation value of various software and hardware testing measures in the next round of testing, select the combination of testing measures with high implementation value based on the calculated implementation value, and generate a maintenance plan. The maintenance plan can include specific testing measures, testing time, and testing personnel arrangements.

[0043] By evaluating the implementation value of testing measures, the most effective testing measures can be selected with limited maintenance resources, thereby improving maintenance efficiency and effectiveness. The maintenance plan generated based on the potential risk parameters of each functional unit of the lithium iron phosphate battery is targeted and can better solve potential problems of the battery, ensuring the stable operation of the lithium iron phosphate battery.

[0044] In one embodiment of the present invention, the step of monitoring the lithium iron phosphate battery and communication module of a communication base station to obtain raw monitoring data includes: S11: Collect data on several battery performance items of the lithium iron phosphate battery in the communication base station to obtain various battery performance data of the lithium iron phosphate battery. S12: Collect data on several communication quality items of the communication module of the communication base station to obtain various communication quality data of the communication module; S13: Combine the battery performance data and the communication quality data to obtain the raw monitoring data.

[0045] Various sensors, such as voltage sensors, current sensors, temperature sensors, and humidity sensors, are installed at key locations in the lithium iron phosphate battery. These sensors must be correctly connected to the data acquisition equipment to ensure accurate data transmission. For example, a voltage sensor is connected in parallel across the battery terminals to measure the battery voltage in real time; a current sensor is connected in series in the circuit to monitor the current. The data acquisition frequency is set appropriately based on the battery's usage characteristics and monitoring needs. For parameters that change rapidly, such as current, a higher acquisition frequency (e.g., once per second) can be set; for parameters that change relatively slowly, such as temperature, the acquisition frequency can be appropriately reduced (e.g., once per minute). The data acquisition equipment transmits the collected battery performance data to a data storage server via wired (e.g., Ethernet, RS-485) or wireless (e.g., Wi-Fi, ZigBee) communication methods. The server stores the data in a database for subsequent analysis and processing.

[0046] Battery parameters such as voltage, current, and temperature are key indicators reflecting its health status and performance. By collecting this data, the battery's working status can be monitored in real time to determine whether the battery is operating normally and whether there are abnormalities such as overcharging, over-discharging, or overheating. Long-term monitoring of various battery performance data can analyze the trend of battery performance changes and predict the remaining battery life. This helps to plan battery replacement in advance and avoid communication base station service interruptions due to battery failure.

[0047] By utilizing specialized network monitoring software and hardware, the communication quality of the communication module is monitored. For example, network analyzers are used to measure signal strength, signal quality, and bit error rate; traffic monitoring tools are used to statistically analyze data transmission rate and packet loss rate. The communication module typically records its own operational logs, including connection establishment and disconnection times, and error messages during communication. Regularly collecting and analyzing this log data provides relevant communication quality information. A remote monitoring system allows for real-time access to various communication quality data from the module. When data anomalies occur, the system can promptly issue alarms and notify maintenance personnel for handling.

[0048] Communication quality directly affects the service quality of communication base stations. Parameters such as signal strength, bit error rate, and packet loss rate reflect the reliability of the communication link. By collecting this data, communication faults and interference sources can be detected in a timely manner, and corresponding measures can be taken for repair and optimization to ensure stable communication operation. Long-term monitoring of communication quality data can assess the performance changes of communication modules. If the communication quality gradually declines, it means that the communication module is aging or malfunctioning, and timely maintenance or replacement is required.

[0049] The collected battery performance data and communication quality data are processed to ensure that the data have the same data type and encoding method. For example, all data are converted into standard timestamp and numerical formats. Based on the data collection time, the battery performance data and communication quality data are correlated and integrated. The database's correlation query function can be used to combine battery data and communication data at the same time point to form a complete original monitoring data record. The integrated data is cleaned and preprocessed to remove invalid data, duplicate data, and outliers. At the same time, the data is normalized and standardized to facilitate subsequent data analysis and modeling.

[0050] Individual battery performance data or communication quality data can only reflect one aspect of the communication base station's condition. Combining the two allows for a comprehensive analysis of the base station's operating status from multiple perspectives, revealing potential correlations between battery performance and communication quality. For example, a decline in battery performance can lead to unstable power supply to the communication module, thus affecting communication quality. The integrated raw monitoring data contains richer information, providing a more comprehensive basis for subsequent data analysis and modeling. Through the analysis of comprehensive data, we can more accurately predict communication base station failures and potential risks, and formulate more effective maintenance strategies.

[0051] In one embodiment of the present invention, the step of assessing the health status of the lithium iron phosphate battery using a pre-trained health supervision model to obtain health assessment information includes: S21: Arrange the raw monitoring data at each time point in time sequence to obtain the raw monitoring sequence of the lithium iron phosphate battery; S22: Determine the information reference mode based on the time characteristics of the current moment, and perform sequence truncation of the original monitoring sequence according to the information reference mode to obtain several segments of health reference information; S23: Substitute the health reference information of each segment into the pre-trained health supervision model to assess the health status of the lithium iron phosphate battery based on the health reference information of each segment, and obtain several health assessment features of the lithium iron phosphate battery. S24: Perform interactive information verification on each of the health assessment features to fuse the information of each of the health assessment features and obtain health assessment information.

[0052] Extract all data records from the database storing the raw monitoring data and identify the timestamp field in each record. The timestamp can be specific date and time information, such as "2024-10-01 12:30:00". Sort the data in ascending order using a suitable sorting algorithm. Common sorting algorithms include bubble sort and quick sort. However, in practical applications, database systems usually provide efficient sorting functions, such as the ORDER BY clause in SQL statements. For example, using SELECT * FROM monitoring_data ORDER BY timestamp ASC can sort the data in the monitoring_data table in ascending order by timestamp. Store the sorted data as a sequence structure, such as an array, list, or data frame. In Python, the pandas library can be used to store the sorted data as a DataFrame object for convenient subsequent processing.

[0053] The operating status of lithium iron phosphate batteries changes continuously over time. By arranging the data in time sequence, we can clearly see the state of the battery at different points in time and discover the evolution of its performance over time, such as the degradation trend of battery capacity. Subsequent operations such as curve plotting and model analysis usually require data to have a clear time sequence. Time sequence data can ensure that the analysis results conform to the actual time logic and facilitate the mining of time-dependent features in the data.

[0054] Analyze the time attributes of the current moment, such as whether it is a weekday or holiday, daytime or nighttime, and which season it is. These time characteristics will affect the battery usage pattern. For example, during the daytime on a weekday, the communication base station has a large traffic volume, and the battery charges and discharges frequently. Predefine information reference patterns corresponding to different time characteristics. For example, for the daytime on a weekday, the reference pattern can be to extract data within 2 hours before and after the current moment; for the nighttime on a holiday, extract data within 4 hours before and after. According to the determined reference pattern, find the corresponding start and end positions in the original monitoring sequence and extract the sequence of specified specifications.

[0055] Batteries perform differently in different time scenarios. Considering the time characteristics of the current moment can make the selected data more representative and closer to the actual use of the battery. This avoids data deviation caused by not considering the time factor and improves the accuracy of health assessment. Extracting sequences of different specifications can provide health reference information at multiple time scales, reflecting the health status of the battery from different perspectives and helping to understand the changes in battery performance more comprehensively.

[0056] The pre-trained health supervision model is loaded from storage media into memory. The model can be trained using machine learning (e.g., decision trees, support vector machines) or deep learning (e.g., neural networks) algorithms. Before inputting health reference information into the model, the data needs to be preprocessed, such as normalized or standardized, to meet the model's input requirements. For example, the StandardScaler function in the sklearn library can be used to standardize the data. The preprocessed health reference information is then input into the loaded model, which calculates based on its internal algorithms and parameters, outputting several health assessment features, such as battery health score and failure probability.

[0057] The pre-trained health supervision model learns the complex relationship between battery data and health status through a large amount of historical data. It can make accurate assessments based on the input health reference information, avoiding the subjectivity and limitations of manual assessment. The health assessment features present the battery's health status in a quantitative form, making the assessment results more intuitive and clear, which facilitates subsequent analysis and decision-making. For example, it can quickly determine whether the battery needs maintenance based on the health score.

[0058] Calculate the correlation coefficient between different health assessment features, such as using the Pearson correlation coefficient. Correlation analysis can help determine the degree of association between features. If two features are too highly correlated, there is redundant information. Check whether each health assessment feature is logical and consistent with the actual situation. For example, the battery health score and failure probability should be within a reasonable range and be consistent with each other. Further investigate and correct feature values ​​that do not conform to logic. Based on the results of correlation analysis and consistency checks, select an appropriate information fusion method, such as the weighted average method or the D-S evidence theory. For example, for features with low correlation and good consistency, the weighted average method can be used for fusion, assigning appropriate weights to each feature.

[0059] Since different health assessment features are obtained from different perspectives or using different algorithms, there are certain errors and uncertainties. Through interactive information verification and fusion, the advantages of each feature can be combined to reduce errors and improve the accuracy and reliability of health assessment information. The health assessment information after information fusion can more comprehensively reflect the health status of lithium iron phosphate batteries, providing stronger support for subsequent maintenance decisions and avoiding one-sided judgments based on a single feature.

[0060] In one embodiment of the present invention, the steps of plotting the health change curve of the lithium iron phosphate battery based on health assessment information at each time point, and marking suspicious points and speculating on the health change curve to generate several speculative health curves include: S31: Map the health assessment information to a designated information space, so as to provide corresponding feedback display on the health assessment information at each information level of the information space, and generate health feedback parameters at each information level; S32: Connect and process the health feedback parameters at each time point of each information level to obtain the hierarchical feedback curve of each information level, which together serve as the health change curve of the lithium iron phosphate battery. S33: Perform parallel analysis on each of the hierarchical feedback curves contained in the health change curve to find feedback doubts on each of the hierarchical feedback curves, and mark the feedback doubts on the hierarchical feedback curves. S34: Based on the feedback points of doubt, the health change curves are adaptively modified to obtain several inferred health curves.

[0061] First, clearly define the specified information space, which can be a multi-dimensional space, such as including time dimension, battery health dimension, and different performance index dimensions (such as charging efficiency, discharge capacity, etc.). Determine the value range and scale of each dimension. Based on the dimensions of the information space and the characteristics of the health assessment information, formulate mapping rules. For example, map the time information in the health assessment information to the time dimension, and map the battery health value to the health dimension. For some complex health assessment indicators, linear transformation, normalization, and other methods can be used to map them to appropriate dimensional intervals. At each information level of the information space, display the health assessment information according to the mapping rules. Visualization tools (such as 3D graphics, heat maps, etc.) can be used to present the health assessment information intuitively. At the same time, based on the display results, generate health feedback parameters for each information level. These parameters can be specific values, levels, or other quantitative representations.

[0062] The multidimensional nature of the information space can display the health status of the battery from multiple perspectives, helping maintenance personnel to have a more comprehensive understanding of the battery's performance in different aspects. For example, by combining the time dimension and different performance index dimensions, the changing trend of battery performance over time and the correlation between various indicators can be observed. The generated health feedback parameters are the basic data for subsequent plotting of hierarchical feedback curves, enabling the curves to accurately reflect changes in the battery's health status.

[0063] The health feedback parameters at each information level are sorted chronologically to ensure data continuity. Sorting algorithms (such as quicksort and mergesort) can be used, or database sorting functions or programming language sorting functions can be utilized. Appropriate curve fitting methods, such as linear fitting, polynomial fitting, and spline fitting, are then used to fit the sorted health feedback parameters. The appropriate fitting method is selected based on the data characteristics and trends to obtain a smooth curve that accurately reflects data changes. For example, if the data shows a linear trend, linear fitting can be used; if the data changes are more complex, polynomial fitting or spline fitting can be used. The feedback curves from each information level are integrated to form the health change curve of the lithium iron phosphate battery. Visualization tools can be used to plot the curves from different levels on the same coordinate system, visually displaying the changes in battery health over time.

[0064] Health change curves can clearly show the trend of battery health over a period of time, enabling maintenance personnel to quickly understand the battery's performance trajectory. For example, by observing the upward or downward trend of the curve, it can be determined whether the battery's health is improving or deteriorating. Hierarchical feedback curves at different information levels can be compared and analyzed, which helps to discover the performance differences and interrelationships of the battery in different aspects. For example, by comparing the hierarchical feedback curves of battery capacity and temperature, the impact of temperature on battery capacity can be analyzed.

[0065] The feedback curves at each level are analyzed to determine their trend, such as rising, falling, or stable. Simultaneously, the correlation between the trends of each curve is analyzed, for example, whether some curves change synchronously or in opposite directions. The trend of the curve can be quantified by calculating indicators such as the slope and curvature. Based on the trend of each part of the hierarchical feedback curve, statistical analysis methods (such as mean, standard deviation, and threshold judgment) are used to detect anomalies in the curves and obtain the first anomaly parameter. For example, if the curve value at a certain moment deviates from the mean by more than a certain standard deviation, then that point is considered to be potentially abnormal.

[0066] Based on the correlation characteristics between the changing trends of each curve, anomaly detection processing is performed on the interaction between the feedback curves at each level to obtain the second anomaly parameter. For example, if two originally related curves suddenly show divergent changing trends, there may be an anomaly. Combining the first anomaly parameter and the second anomaly parameter, the feedback curves at each level are judged for suspicious points, and the suspicious points are identified. Specific symbols (such as asterisks, circles, etc.) or colors are used to mark these suspicious points on the hierarchical feedback curves to facilitate identification and subsequent analysis.

[0067] By analyzing the curves side-by-side and detecting anomalies, abnormal points in the battery health status can be identified. These anomalies indicate potential problems with the battery, such as battery aging or potential malfunctions. Timely detection of these problems can provide important clues for subsequent maintenance work. Marking the reported suspicious points allows maintenance personnel to quickly locate the problem, avoiding a comprehensive inspection of the entire curve and improving the efficiency of maintenance work.

[0068] Using the feedback point of concern as a benchmark, historical and future information before and after the point of concern is selected on the health change curve. These serve as historical and developmental reference information for the feedback point of concern. The time range of the selected information can be determined according to actual needs. For example, data from 10 time points before and 10 time points after the point of concern can be selected. Based on the historical reference information, statistical analysis methods or machine learning models are used to generate several inferred feedback parameters for the feedback point of concern. For example, the reasonable value range of the point of concern can be predicted based on the changing patterns of historical data. Alternatively, time series prediction models (such as ARIMA, LSTM, etc.) can be used to generate inferred feedback parameters. These inferred feedback parameters can be used to replace the feedback point of concern, and the future development of the inferred feedback parameters can be simulated based on the developmental reference information. The same curve fitting method or time series prediction model can be used to simulate the future development trend of battery health status, thereby obtaining the inferred health curves for each inferred feedback parameter.

[0069] Inferred health curves can help maintenance personnel predict the future health status of batteries and prepare for maintenance in advance. For example, if the inferred health curve shows that the battery health status will deteriorate rapidly, then battery replacement or repair can be arranged in a timely manner. Generating several inferred health curves can take into account different possibilities and provide multiple reference options for maintenance decisions. Maintenance personnel can formulate corresponding maintenance strategies based on the different inferred curves to reduce maintenance risks.

[0070] In one embodiment of the present invention, the step of performing parallel analysis on each of the hierarchical feedback curves included in the health change curve to find feedback discrepancies on each of the hierarchical feedback curves includes: S331: Perform parallel analysis on each of the hierarchical feedback curves contained in the health change curve to obtain the curve change trend of each hierarchical feedback curve and the correlation characteristics between the curve change trends; S332: Based on the curve change trend of each part of the hierarchical feedback curve, perform anomaly detection processing on the hierarchical feedback curve itself to obtain the first anomaly parameter of each part of the hierarchical feedback curve. S333: Based on the correlation characteristics between the changing trends of each curve, perform anomaly detection processing on the interaction between each hierarchical feedback curve to obtain the second anomaly parameter of each part of each hierarchical feedback curve. S334: Combine the first abnormal parameter and the second abnormal parameter to determine the suspicious points on each level of feedback curve, so as to identify the suspicious points on the level of feedback curve.

[0071] The feedback curves at each level are plotted on the same coordinate system and distinguished by different colors or line styles to visually observe the trend of each curve. For each level of feedback curve, its slope, curvature, and other indicators are calculated to quantify the curve's changing trend. For example, the least squares method is used to fit local segments of the curve, and the slope of the fitted line is used to determine the upward or downward trend of the curve in that local area. The correlation coefficient (such as the Pearson correlation coefficient) between different levels of feedback curves is calculated to analyze the synchronicity or inverse relationship between the curve's changing trends. Furthermore, methods such as the Granger causality test can be used to determine whether there is a causal relationship between the curves.

[0072] By parallel analysis and trend calculation, we can clearly grasp the changing patterns of each level of feedback curve, such as whether it shows an upward, downward, or fluctuating trend, and what the rate of change is. Analyzing the correlation characteristics between the changing trends of the curves can reveal the mutual influence between the battery performance indicators represented by different levels of feedback curves, providing an important basis for subsequent anomaly detection.

[0073] Based on historical data of the curve, its mean, standard deviation, and other statistical quantities are calculated. A threshold for the normal fluctuation range is set; for example, the standard deviation plus or minus a certain multiple of the mean can be considered the normal range. Data points exceeding this range are considered abnormal. To detect whether the slope or curvature of the curve changes abruptly at a certain moment, the difference in slope or curvature between adjacent time periods can be calculated. When the difference exceeds a set threshold, that point is determined to be a trend inflection point. For detected abnormal points, a first anomaly parameter is calculated based on factors such as the degree of deviation from the normal range and the magnitude of the trend inflection. The greater the degree of anomaly, the higher the value of the first anomaly parameter.

[0074] Each level of feedback curve has its own normal variation pattern. By detecting its own anomalies, points in the curve that do not conform to the normal pattern can be found. These points indicate that the battery has an abnormality in this indicator. The first anomaly parameter can quantify the anomaly, which is convenient for subsequent comprehensive judgment and comparison of the severity of anomalies at different locations.

[0075] Based on the correlation characteristics between the curve trends obtained from the previous analysis, we can verify whether the feedback curves at each level conform to these correlation rules at each time point. For example, if two curves are usually positively correlated, when one curve rises while the other curve falls, it is judged as an anomaly. For cases that do not conform to the correlation rules, we can calculate the degree of abnormal correlation. This can be calculated based on factors such as the change in the correlation coefficient between the curves and the degree of deviation from the normal correlation pattern. The higher the degree of abnormal correlation, the larger the value of the second anomaly parameter.

[0076] Different performance indicators of a battery are often interrelated. When the correlation between curves is broken, it means that there is a potential fault or abnormality in the battery. Such correlation anomalies can be discovered through interactive anomaly detection. Some anomalies may be difficult to detect in the anomaly detection of a single curve, but can be revealed through interactive detection between curves, thus more comprehensively detecting abnormalities in the battery.

[0077] Assign appropriate weights to the first and second anomaly parameters, and then sum them up to obtain the comprehensive anomaly parameter. The weights can be adjusted according to the actual situation. For example, if more attention is paid to the anomaly of the curve itself, the first anomaly parameter can be given a higher weight. Based on historical data or experience of the comprehensive anomaly parameter, set a threshold for suspicious points. When the comprehensive anomaly parameter of a certain data point exceeds the threshold, the point is determined to be a feedback suspicious point. Mark the identified feedback suspicious points on the hierarchical feedback curve using special symbols or colors to facilitate subsequent viewing and analysis.

[0078] The first and second abnormal parameters reflect the abnormal situation from the perspectives of the curve itself and the correlation between curves, respectively. By combining the two, feedback points can be identified more comprehensively and accurately. After identifying feedback points, maintenance personnel can focus their attention on these potentially problematic points, thereby improving the efficiency and pertinence of maintenance work.

[0079] In one embodiment of the present invention, the step of adaptively correcting the health change curves based on the feedback points to obtain several inferred health curves includes: S341: Based on the feedback doubts, select historical and future information on the health change curve to obtain historical and developmental reference information of the feedback doubts. S342: Based on the historical reference information, generate several speculative feedback parameters for the feedback doubts, and use the speculative feedback parameters to replace the feedback doubts, so as to simulate the future development of the speculative feedback parameters according to the development reference information, so as to obtain the speculative health curves corresponding to each of the speculative feedback parameters.

[0080] Based on the corresponding time points on the health change curve for the feedback issues, determine the time range for selecting historical and future information. This time range can be adjusted according to the actual situation. For example, select data from the 24 hours before the feedback issue as historical reference information and data from the 12 hours after the feedback issue as development reference information. On the health change curve, filter out the corresponding data points according to the determined time range. Data can be filtered by timestamps to ensure that the selected data is within the specified time interval. Organize the filtered historical and future data points into historical reference information and development reference information, respectively. This information can be stored in the form of arrays, lists, or data frames for convenient subsequent processing.

[0081] Historical reference information can reflect the normal or abnormal development pattern of the battery before the emergence of feedback doubts, providing a basis for generating inferred feedback parameters. Development reference information helps to simulate the possible development trend of the battery after the feedback doubts, making the inference more consistent with the actual situation. By analyzing historical and future information, we can capture the changing pattern of battery health status over time, thereby better understanding the reasons for the emergence of feedback doubts and their possible impact.

[0082] Based on the statistical characteristics of historical reference information, such as mean, median, and standard deviation, inferred feedback parameters can be generated. For example, assuming the mean of historical data is μ and the standard deviation is σ, multiple inferred feedback parameters within the range of μ ± kσ (k is a constant) can be generated. Using time series prediction models (such as ARIMA, LSTM, etc.), the possible values ​​of feedback points can be predicted based on historical reference information, generating multiple different inferred feedback parameters. By adjusting the model parameters or inputting different initial conditions, multiple different prediction results can be obtained. Each generated inferred feedback parameter can be used to replace the feedback point on the health change curve to obtain multiple modified health change curves.

[0083] The health change curve after replacement is trained using development reference information. Then, the trained time series model is used to predict the battery health status in the future period to obtain the future development trend of each inferred feedback parameter. Based on the trend of development reference information, the health change curve after replacement is fitted (such as linear fitting, polynomial fitting, etc.). Then, the fitted curve is extrapolated to the future time to obtain the inferred health curve.

[0084] Since the emergence of feedback doubts can have multiple causes and development directions, generating multiple speculative feedback parameters can take into account different possibilities, avoid the limitations of a single prediction, and predict the future health status of the battery under different conditions through future development simulation. This provides maintenance personnel with multiple possible scenarios, helping them to formulate response strategies in advance and reduce potential risks. For example, if some speculative health curves show that the battery health status will deteriorate sharply, maintenance measures such as battery replacement can be arranged in a timely manner.

[0085] In one embodiment of the present invention, the step of assessing the potential risks of the lithium iron phosphate battery by combining the health change curves and the various inferred health curves, and allocating corresponding maintenance plans for the lithium iron phosphate battery based on the assessment results, includes: S41: Extract the curve difference features between each of the predicted health curves and the health change curves, and express the curve difference features in a vectorized form to obtain the difference feature matrix; S42: Based on the difference feature matrix, the operating status of each functional unit of the lithium iron phosphate battery is predicted to obtain the operating prediction information of each functional unit of the lithium iron phosphate battery. S43: Perform cross-validation on each of the operational speculation information to obtain the potential risk parameters of each functional unit of the lithium iron phosphate battery; S44: Retrieve the test records of the lithium iron phosphate battery to schedule the next round of testing for the lithium iron phosphate battery; S45: Based on the potential risk parameters, evaluate the implementation value of several existing software and hardware testing measures to obtain the implementation value of each software and hardware testing measure in the next round of testing, and generate a maintenance plan based on each implementation value.

[0086] At the same time points, the numerical differences between corresponding points on the predicted health curve and the health change curve are calculated, such as differences and ratios, and the differences in shape characteristics such as slope and curvature of the curves are analyzed. For example, the difference in slope of the two curves at each time point is calculated, and the consistency of the upward, downward, and stable trends of the curves is compared. If they are inconsistent, this trend difference is quantified, and the extracted curve difference features are arranged into a vector in a certain order. For example, the numerical differences and slope differences at each time point are arranged sequentially. The difference feature vectors of each predicted health curve and the health change curve are combined to form a difference feature matrix, where each row of the matrix corresponds to a difference feature vector of the predicted health curve.

[0087] By extracting curve difference features and expressing them quantitatively, the differences between curves are transformed into specific values, which facilitates subsequent calculations and analysis. The difference feature matrix can comprehensively reflect the differences between each inferred health curve and the health change curve. These difference features contain potential operational change information of lithium iron phosphate batteries and are an important basis for judging the potential risks of lithium iron phosphate batteries.

[0088] Based on historical data and domain knowledge, a correlation model can be established between the difference feature matrix and the operating status of each functional unit of the battery. Machine learning algorithms, such as decision trees and neural networks, can be used, or rule-based methods can be adopted. The difference feature matrix is ​​input into the correlation model, and the model infers based on the input features and outputs operational prediction information of each functional unit of the lithium iron phosphate battery, such as whether the charging module is normal and whether the efficiency of the discharging module has decreased.

[0089] By delving into the differences in the curves to understand the operating status of each functional unit of the battery, it is possible to more accurately locate the parts of the battery that may have problems, providing a more specific direction for subsequent maintenance. Predicting the operating status of each functional unit can help identify potential fault hazards in advance and prevent the fault from developing further and causing the lithium iron phosphate battery to fail.

[0090] Collect other relevant information besides operational projections, such as historical battery fault records and environmental monitoring data. Integrate this multi-source information with the operational projections. Consistency verification can be used to check whether information from different sources is consistent; logical verification can also be used to check whether the operational projections conform to the battery's working principles and logical rules. Based on the results of cross-validation, calculate the potential risk parameters for each functional unit of the lithium iron phosphate battery. For example, a risk score can be assigned to each functional unit as a potential risk parameter based on factors such as the degree of consistency of information and the severity of abnormal situations.

[0091] Single operational predictions may contain errors or uncertainties. Cross-validation can integrate information from multiple sources, reduce errors, and improve the accuracy of potential risk assessment. Accurate potential risk parameters can help maintenance personnel develop maintenance plans more scientifically and allocate maintenance resources more rationally.

[0092] Retrieve testing records of lithium iron phosphate batteries from databases or other storage systems, including previous testing times and results. Based on factors such as battery type, usage, and potential risk parameters, formulate rules for scheduling the next round of testing. For example, for lithium iron phosphate batteries with higher potential risks, the testing cycle can be shortened; for lithium iron phosphate batteries in good operating condition, the testing cycle can be appropriately extended. Determine the next round of testing time for lithium iron phosphate batteries based on the testing records and the established rules.

[0093] By retrieving test records and scheduling the next round of testing in conjunction with the actual condition of the battery, over-testing or under-testing can be avoided. A reasonable testing plan can reduce testing costs while ensuring the safe operation of the battery. Regular testing can promptly identify new problems that arise during battery operation, ensuring that lithium iron phosphate batteries are always in good operating condition.

[0094] Establish an implementation value assessment model for software and hardware testing measures. This model can consider factors such as potential risk parameters, testing costs, and testing effects. For example, for functional units with high potential risks, testing measures with good testing effects but high costs have high implementation value. Input the potential risk parameters of each functional unit into the assessment model, calculate the implementation value of various software and hardware testing measures in the next round of testing, select the combination of testing measures with high implementation value based on the calculated implementation value, and generate a maintenance plan. The maintenance plan can include specific testing measures, testing time, and testing personnel arrangements.

[0095] By evaluating the implementation value of testing measures, the most effective testing measures can be selected with limited maintenance resources, thereby improving maintenance efficiency and effectiveness. The maintenance plan generated based on the potential risk parameters of each functional unit of the lithium iron phosphate battery is targeted and can better solve potential problems of the battery, ensuring the stable operation of the lithium iron phosphate battery.

[0096] like Figure 2 As shown, this invention provides an artificial intelligence-based predictive maintenance system for communication base station batteries, used to implement the artificial intelligence-based predictive maintenance method for communication base station batteries as described in any one of the first aspects, comprising: The data monitoring module is used to monitor the lithium iron phosphate battery and communication module of the communication base station to obtain raw monitoring data. The health assessment module is used to assess the health status of the lithium iron phosphate battery based on the raw monitoring data using a pre-trained health supervision model, and to obtain health assessment information. The suspicious point prediction module is used to draw the health change curve of the lithium iron phosphate battery based on the health assessment information at each time point, and to mark suspicious points and predict the condition of the health change curve to generate several predicted health curves. The risk maintenance module is used to assess the potential risks of the lithium iron phosphate battery by combining the health change curves and the inferred health curves, and to assign a corresponding maintenance plan to the lithium iron phosphate battery based on the assessment results.

[0097] In this embodiment, the specific implementation of each module in the above system embodiment is described in the above method embodiment, and will not be repeated here.

[0098] The above description is merely a specific embodiment of the invention, but the scope of protection of the invention is not limited thereto. Any changes or substitutions conceived without creative effort should be included within the scope of protection of the invention.

Claims

1. A predictive maintenance method for communication base station batteries based on artificial intelligence, characterized in that, include: Data monitoring is performed on the lithium iron phosphate batteries of communication base stations to obtain raw monitoring data; The health status of the lithium iron phosphate battery is assessed by a pre-trained health supervision model based on the raw monitoring data, and health assessment information is obtained. Based on the health assessment information at each time point, the health change curve of the lithium iron phosphate battery is plotted, and the suspicious points of the health change curve are marked and the condition is inferred to generate several inferred health curves. The potential risks of the lithium iron phosphate battery are assessed by combining the health change curves and the inferred health curves, and corresponding maintenance plans are assigned to the lithium iron phosphate battery based on the assessment results.

2. The predictive maintenance method for communication base station batteries based on artificial intelligence as described in claim 1, characterized in that, The steps for monitoring the lithium iron phosphate batteries in communication base stations to obtain raw monitoring data include: Data on several battery performance items of the lithium iron phosphate battery in the communication base station were collected to obtain various battery performance data of the lithium iron phosphate battery. Data on several communication quality items are collected from the communication module of the communication base station to obtain various communication quality data of the communication module; The battery performance data and the communication quality data are combined to obtain the raw monitoring data.

3. The predictive maintenance method for communication base station batteries based on artificial intelligence as described in claim 1, characterized in that, The steps for assessing the health status of lithium iron phosphate batteries using a pre-trained health supervision model to obtain health assessment information include: The raw monitoring data at each time point are arranged in time sequence to obtain the raw monitoring sequence of the lithium iron phosphate battery; Based on the time characteristics of the current moment, an information reference pattern is determined, and the original monitoring sequence is truncated in several specifications according to the information reference pattern to obtain several segments of health reference information. The health reference information of each segment is substituted into a pre-trained health supervision model to assess the health status of the lithium iron phosphate battery based on the health reference information of each segment, thereby obtaining several health assessment features of the lithium iron phosphate battery. Interactive information verification is performed on each of the health assessment features to fuse the information of each health assessment feature and obtain health assessment information; Among them, the pre-trained health supervision model is a multi-model combination structure, including a rule-based judgment model based on battery safety operation thresholds to identify obvious abnormalities such as overvoltage, overcurrent, and overtemperature; a support vector machine (SVM) for small sample health status classification; a random forest for multi-feature health scoring and importance analysis; and an LSTM for short-term trend prediction under stable operating conditions.

4. The predictive maintenance method for communication base station batteries based on artificial intelligence as described in claim 1, characterized in that, The steps of plotting the health change curve of the lithium iron phosphate battery based on health assessment information at each time point, marking suspicious points and speculating on the health change curve, and generating several speculative health curves include: The health assessment information is mapped to a designated information space, so that the health assessment information is displayed in a corresponding manner at each information level of the information space, and health feedback parameters are generated at each information level. The health feedback parameters at each time point of each information level are connected and processed to obtain the hierarchical feedback curves of each information level, which together serve as the health change curve of the lithium iron phosphate battery. Parallel analysis is performed on each of the hierarchical feedback curves included in the health change curve to find feedback discrepancies on each hierarchical feedback curve, and the feedback discrepancies are marked on the hierarchical feedback curve. Based on the feedback points of doubt, the health change curves are adaptively modified to obtain several inferred health curves.

5. The predictive maintenance method for communication base station batteries based on artificial intelligence as described in claim 4, characterized in that, The step of performing a parallel analysis of the hierarchical feedback curves contained in the health change curves to identify feedback discrepancies on each hierarchical feedback curve includes: The hierarchical feedback curves contained in the health change curve are analyzed in parallel to obtain the curve change trend of each hierarchical feedback curve and the correlation characteristics between the curve change trends. Based on the curve change trends of each part of the hierarchical feedback curve, the hierarchical feedback curve itself is subjected to anomaly detection processing to obtain the first anomaly parameter of each part of the hierarchical feedback curve. Based on the correlation characteristics between the changing trends of each curve, anomaly detection processing of the interaction between each hierarchical feedback curve is performed to obtain the second anomaly parameter of each part of each hierarchical feedback curve. By combining the first abnormal parameter and the second abnormal parameter, the feedback curves of each level are judged to identify the feedback suspicious points on the feedback curves.

6. The predictive maintenance method for communication base station batteries based on artificial intelligence as described in claim 4, characterized in that, The steps for adaptively correcting the health change curves based on the feedback points to obtain several inferred health curves include: Based on the feedback doubts, historical and future information are selected on the health change curve to obtain historical and developmental reference information of the feedback doubts. Based on the historical reference information, several speculative feedback parameters are generated for the feedback doubts, and the speculative feedback parameters are used to replace the feedback doubts. The future development of the speculative feedback parameters is simulated according to the development reference information to obtain the speculative health curves for each of the speculative feedback parameters.

7. The predictive maintenance method for communication base station batteries based on artificial intelligence as described in claim 1, characterized in that, The steps of assessing the potential risks of the lithium iron phosphate battery by combining the health change curves with the inferred health curves, and allocating corresponding maintenance plans for the lithium iron phosphate battery based on the assessment results, include: The curve difference features of each of the predicted health curves and the health change curves are extracted and expressed in vectorized form to obtain the difference feature matrix; Based on the difference feature matrix, the operating status of each functional unit of the lithium iron phosphate battery is inferred, and the operating inference information of each functional unit of the lithium iron phosphate battery is obtained. Cross-validation is performed on each of the operational speculation information to obtain the potential risk parameters of each functional unit of the lithium iron phosphate battery; Retrieve the testing records of the lithium iron phosphate battery to schedule the next round of testing for the lithium iron phosphate battery; Based on the aforementioned potential risk parameters, the implementation value of several existing software and hardware testing measures is evaluated to obtain the implementation value of each software and hardware testing measure in the next round of testing, so as to generate a maintenance plan based on each implementation value.

8. A predictive maintenance system for communication base station batteries based on artificial intelligence, characterized in that, A method for implementing the predictive maintenance method for communication base station batteries based on artificial intelligence as described in any one of claims 1-7 includes: The data monitoring module is used to monitor the lithium iron phosphate battery and communication module of the communication base station to obtain raw monitoring data. The health assessment module is used to assess the health status of the lithium iron phosphate battery based on the raw monitoring data using a pre-trained health supervision model, and to obtain health assessment information. The suspicious point prediction module is used to draw the health change curve of the lithium iron phosphate battery based on the health assessment information at each time point, and to mark suspicious points and predict the condition of the health change curve to generate several predicted health curves. The risk maintenance module is used to assess the potential risks of the lithium iron phosphate battery by combining the health change curves and the inferred health curves, and to assign a corresponding maintenance plan to the lithium iron phosphate battery based on the assessment results.

Images